Publications

A survey on Concept-Based Approaches for Model Improvement
Avani Gupta, P J Narayanan , arxiv, 2024
Paper

Surveyed various concept representation, discovery and concept based model improvement methods.

	Predicting Business Process Events in Presence of Anomalous IT Events Avani Gupta, Avirup Saha , Sambit Ghosh, Neelamadhav Gantayat, Renuka Sindhgatta , CODS-CODMAD, 2024 Paper Predicted possible IT errors (delays/violations/anamolies) in Business Processes.
	Concept Distillation: Leveraging Human-Centered Explanations for Model Improvement Avani Gupta , Saurabh Saini , P J Narayanan , Neurips 2023 Paper \| Project Page \| Code Proposed a novel human centered concept based training for (de)sensitizing models towards concepts.
	Interpreting Intrinsic Image Decomposition using Concept Activations Avani Gupta , Saurabh Saini , P J Narayanan , ICVGIP, 2022 (Oral) *Best Paper Award* Paper \| Project Page \| Code Proposed a novel evaluation stratgy and metric (Concept Sensitivity Metric) for evaluation of ill-posed posed underconstrained problems by measuring disentanglement.
	Goal-Oriented Next Best Activity Recommendation using Reinforcement Learning Prerna Agarwal, Avani Gupta, Renuka Sindhgatta, Sampath Dechu, arxiv, 2022 Paper Modeled Next Best action prediction as a RL problem with changing action spaces.
	CitRet: A Hybrid Model for Cited Text Span Retrieval Amit Pandey, Avani Gupta, Vikram Pudi, Coling, 2022 Paper \| Code A novel model for Text Span Retrieval which is used for summarization of scientific documents.
	Fake news detection using Deep Learning based Natural Language Processing Saanika Gupta, Vinayak Bhartiya, Avani Gupta, Nevnath Srinivas N Equal Contribution Student Research Symposium, HiPC, 2019 Trained an artificial neural network with single hidden layer for Fake news detection. Used credit history of users: score based on what type of news shared by user in past. Trained on Liar’s Dataset, optimized various hyper-parameters and achieved 30% increase in accuracy than baselines.
	Abstract representation of visual stimuli from neural recordings using deep generative models Avani Gupta, Rishabh Chakraborty, * Equal Contribution Poster, Ernst Strüngmann Institutefor Neuroscience Conference (ESI Sync) in Cooperation with Max-Planck-Society, 2020 View abstract Proposed a self-supervised 3D reconstruction pipeline which allows to train an end-to-end fMRI-to-3D object reconstruction using a 3D Generative- Adversarial Modeling technique. Used Encoder-Decoder Architecture; the variational auto-encoder encodes the 3D object (corresponding to the fMRI signal) to fMRI signal. This signal is passed onto the decoder which constructs back the 3D object. Used 3D VAE GAN.
	Reconstruction of the perceived visual stimuli using 3D Generative-Adversarial Modeling Avani Gupta, Rishabh Chakraborty, Karthik Vishvanathan* * Equal Contribution Neuromatch Conference, 2020 Proposed a self-supervised 3D reconstruction pipeline which allows to train an end-to-end fMRI-to-3D object reconstruction using a 3D Generative- Adversarial Modeling technique

	Predicting Business Process Events in Presence of Anomalous IT Events Avani Gupta, Avirup Saha , Sambit Ghosh, Neelamadhav Gantayat, Renuka Sindhgatta , CODS-CODMAD, 2024 Paper Predicted possible IT errors (delays/violations/anamolies) in Business Processes.
	Concept Distillation: Leveraging Human-Centered Explanations for Model Improvement Avani Gupta , Saurabh Saini , P J Narayanan , Neurips 2023 Paper \| Project Page \| Code Proposed a novel human centered concept based training for (de)sensitizing models towards concepts.
	Interpreting Intrinsic Image Decomposition using Concept Activations Avani Gupta , Saurabh Saini , P J Narayanan , ICVGIP, 2022 (Oral) *Best Paper Award* Paper \| Project Page \| Code Proposed a novel evaluation stratgy and metric (Concept Sensitivity Metric) for evaluation of ill-posed posed underconstrained problems by measuring disentanglement.
	Goal-Oriented Next Best Activity Recommendation using Reinforcement Learning Prerna Agarwal, Avani Gupta, Renuka Sindhgatta, Sampath Dechu, arxiv, 2022 Paper Modeled Next Best action prediction as a RL problem with changing action spaces.
	CitRet: A Hybrid Model for Cited Text Span Retrieval Amit Pandey, Avani Gupta, Vikram Pudi, Coling, 2022 Paper \| Code A novel model for Text Span Retrieval which is used for summarization of scientific documents.
	Fake news detection using Deep Learning based Natural Language Processing Saanika Gupta, Vinayak Bhartiya, Avani Gupta, Nevnath Srinivas N Equal Contribution Student Research Symposium, HiPC, 2019 Trained an artificial neural network with single hidden layer for Fake news detection. Used credit history of users: score based on what type of news shared by user in past. Trained on Liar’s Dataset, optimized various hyper-parameters and achieved 30% increase in accuracy than baselines.
	Abstract representation of visual stimuli from neural recordings using deep generative models Avani Gupta, Rishabh Chakraborty, * Equal Contribution Poster, Ernst Strüngmann Institutefor Neuroscience Conference (ESI Sync) in Cooperation with Max-Planck-Society, 2020 View abstract Proposed a self-supervised 3D reconstruction pipeline which allows to train an end-to-end fMRI-to-3D object reconstruction using a 3D Generative- Adversarial Modeling technique. Used Encoder-Decoder Architecture; the variational auto-encoder encodes the 3D object (corresponding to the fMRI signal) to fMRI signal. This signal is passed onto the decoder which constructs back the 3D object. Used 3D VAE GAN.
	Reconstruction of the perceived visual stimuli using 3D Generative-Adversarial Modeling Avani Gupta, Rishabh Chakraborty, Karthik Vishvanathan* * Equal Contribution Neuromatch Conference, 2020 Proposed a self-supervised 3D reconstruction pipeline which allows to train an end-to-end fMRI-to-3D object reconstruction using a 3D Generative- Adversarial Modeling technique